Abstract
The extended and increasing use of rubber on engineering systems or industrial products, in the last decades, is a fact. Therefore, the products with rubber have a large range of applications in almost any industrial field, e.g. power transmission, oil industry, etc. In the power transmission case, the main function of the rubber is as an electrical insulator. The underground cables are an example where the rubber has an essential importance to guarantee their effective and reliable life time. In the oil industry, rubber could be used in aggressive environments for vibration and noise control or as structural elements. For example, viscoelastic materials are known to act as efficient dampers of structural vibrations, considering their great energy loss hysteresis cycles, which reduce harmful effects of high level vibrations and noise on humans, as well as on machinery. The several environmental conditions around the rubber can induce different kind of degradation processes. Identification and well understanding of these processes constitute a challenge that could reduce the costs of unexpected faults, increasing the reliability of such products. When a degradation process arises, the rubber main properties are modified. Nowadays, in the technical literature available, one can find a large amount of methods that characterize this dynamic behavior which, however, are not related with a degradation process. Hence, the development of methodologies aiming a rapid and precise determination of any degradation process occurring on a particular system is very important. The objective of this paper is to present a new method for the performance characterization and classification of these materials based on behavior parameter, i.e. fundamental frequencies, amplitudes and loss factors, under severe environmental conditions. This would permit to identify degradation processes in course. In order to reach these objectives, the frequency response functions of Oberst beams with a rubber layer (new and degraded) were evaluated (based on ASTM 756-83 specification) and the dynamic parameters were extracted. All data were classified using neural networks based on self-organizing maps, as a way to indentify normal and abnormal "clusters" related with the rubber degradation state. Furthermore, an anomaly detection system was implemented using a MATLAB interface, able to automate and speed up the assessment process. The next step of this work will be the development of an expert diagnostic system in order to determine what kind of degradation process is occurring, according to the information observed and the knowledge available.
Identification of degradation behavior on a rubber material using dynamic parameters and self-organized maps
